Skip to content

Benchmark

BALROG BabyAI

Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages.

Models measured
37
Scores recorded
38
Sources
1
IQ weight
1.20

Score distribution

63.1%
Mean
68%
Median
26.7%
Std dev
24%
10th percentile
98%
90th percentile
4%
Min
4%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.20 for chat models, measured over 37 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 5max· Epoch AI Benchmarking Hub data100%
2Gemini 3.1 Pro Preview· Epoch AI Benchmarking Hub data100%
3GPT-6 Astramax· Epoch AI Benchmarking Hub data100%
4GPT-5.6 Lunamax· Epoch AI Benchmarking Hub data98%
5GPT-5.6 Solmax· Epoch AI Benchmarking Hub data98%
6GPT-5.6 Terramax· Epoch AI Benchmarking Hub data98%
7Gemini 3 Pro Preview· Epoch AI Benchmarking Hub data96%
8Gemini 3 Flash· Epoch AI Benchmarking Hub data86%
9Claude Opus 4.5· Epoch AI Benchmarking Hub data80%
10Gemini 2.5 Pro· Epoch AI Benchmarking Hub data80%
11GPT 5minimal· Epoch AI Benchmarking Hub data80%
12GPT 4o· Epoch AI Benchmarking Hub data77.6%
13Grok 4· Epoch AI Benchmarking Hub data76%
14Reka Flash 3· Epoch AI Benchmarking Hub data76%
15DeepSeek R1· Epoch AI Benchmarking Hub data74%
16Llama 3.1 70B Instruct· Epoch AI Benchmarking Hub data73.2%
17Claude Opus 4.564k· Epoch AI Benchmarking Hub data72%
18Llama 3.2 90B Vision Instruct· Epoch AI Benchmarking Hub data72%
19Claude 3.5 Sonnet· Epoch AI Benchmarking Hub data68%
20Gemini 2.5 Flash· Epoch AI Benchmarking Hub data68%
21Claude Haiku 4.5· Epoch AI Benchmarking Hub data66%
22Llama 3.3 70B Instruct· Epoch AI Benchmarking Hub data66%
23Grok 3beta· Epoch AI Benchmarking Hub data62%
24gemini-1.5-pro-002· Epoch AI Benchmarking Hub data58.4%
25Claude Haiku 3.5· Epoch AI Benchmarking Hub data52%
26GPT-4o mini· Epoch AI Benchmarking Hub data50.4%
27gemini-1.5-flash-002· Epoch AI Benchmarking Hub data50%
28Llama 3.2 11B Vision Instruct· Epoch AI Benchmarking Hub data50%
29Mistral Nemo· Epoch AI Benchmarking Hub data50%
30DeepSeek R1 Distill Qwen 32B· Epoch AI Benchmarking Hub data48%
31Llama 3.1 8B Instruct· Epoch AI Benchmarking Hub data36%
32Qwen2.5 72B Instruct· Epoch AI Benchmarking Hub data34%
33Phi 4· Epoch AI Benchmarking Hub data32%
34Qwen2-VL (72B) Instruct· Epoch AI Benchmarking Hub data24%
35Llama 3.2 3B· Epoch AI Benchmarking Hub data20%
36Qwen2.5 7B Instruct· Epoch AI Benchmarking Hub data14%
37Llama 3.2 1B Instruct· Epoch AI Benchmarking Hub data8%
38Qwen2 VL 7b Instruct· Epoch AI Benchmarking Hub data4%

Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (licence)